I've been busy [and apparently built up quite a sciforums post - sorry about the length of my rant] and seem to have missed an interesting and animated discussion. It seems to me there has been a lot of confusion over language. I'm going to post with the context of the discussion in mind, but not attempt to reply to any one persons comments directly as that would just be confusing ...
Evolution is surprisingly just defined as genetic change of a population (as pointed out by paulsamuel). People often use the term implying some type of progress. As no good measure of "genetic progress" exists change is the only requirement we place on it. The most satisfying aspect of this definition is that without change you can be sure no evolution of any sort or progress is being made. The unsatisfying part is that I highly doubt there are any populations which are balanced such that there is no change in gene frequencies over time (even if they are fairly stable fluctuations are unavoidable).
Is change inherently beneficial in such a way that one can expect it to lead to progress? Again one falters on what exactly to define as the measure for progress. There are so many possibilities I shudder to go down that road. To approach it in a different way one could instead ask the question: is any species perfect? While this approach has the same problem of defining perfection I think most would agree that none is. The one further complication is that our environment is dynamic and what is perfect certainly is context dependent.
In terms of definitions, a gene is considered to have a positive contribution to fitness if its frequency is increasing over time. Again this seems like a bit of a dodge, but the complexity makes any explicit measure beyond this highly arbitrary. At the same time here are documented instances in pairs of alleles whose frequencies oscillate seasonally, particularly in species with life spans under a year. Always evolving, but making no more progress than sysyphus.
At any point in time there are certain organisms which can survive. Certain compositions survive more on average and increase in frequency. The fact that some stick around is important as in the case of the cycling as changes in the environment place can easily bring a gene or set of genes back into vogue. If they were eliminated completely the decline of the one subset would translate into a decline of a whole species (or subpopulation of a species). In the extreme extinction.
So it would seem maintaining a pool of variation is of a benefit to a population. A species however can only increase it's variability to a certain point and then problems of incompatability in certain combinations will begin to limit further variation. Population size also places practical limits on variability. Selection also removes a lot of variability, in the extreme instance dominant lethal genes. Variation is only good in the long run, not in the short. It is the constant change in environment that helps maintain and over time select for organisms that can maintain variation.
I would like to distinguish between two different types of variability - there a specific combinations of genes that are rare and there are certain alleles that are rare (allele = version of a gene). If you mix two populations where the set of alleles in each subpopulation are different then for sure there will be an increase in the total number of allele variants and possible combinations in this mixed population. However the total number of alleles in the species remains unchanged so this increase variability is only relative to the previous two subpopulations. The existence of new combinations that didn't exist before is a clear increase in the existence of specific variants that exist. By the same token a simple exercise in statistics will demonstrate the fact that the number of realized variants that occur in the combined population will decrease relative to that of the two independent populations taken together. Combinations seem to be very important in evolution in higher eukaryotes.
This can basically be explained by thermodynamics. If you have two chambers - one which contains a vacuum and one which contains air separated by a stopper. Once you remove the stopper you will never see either side become vacuum. Similarly if you take a two populations of 100 urns - one that contains 70:30 mix of green:red chips and one that contains the reverse 30:70 green:red. If you take each and mix it each "generation" by pairing up the urns dumping one into the other in each pair then resplitting the chips between them and count the number of variant combinations it will be more than if the populations are combined. While both are very different from the dynamic state of things in biology they hold much of the insight necessary to understand the decrease in variation upon subpopulations mixing.
Regarding small populations and evolution, there are two things that can allow rapid change. First in generating the small population sampling errors or fluctuations occur which can lead to populations with a distinct skew (while one might think on average they should be a good representation there is the clear point that at a certain size sample in which the total variability of a population can't be representation and before this point is reached there will typically be a threshholding loss of variability). Also there is a reduction in inertia which will allow alleles to pass through the population much more quickly (this can be added to by geographic spread of the species as well). While there are many reasons such populations are at risk, if they happen frequently enough in some instances they will make it and the altered circumstances can alter their "evolutionary trajectory" or general dynamics going between the composition of the gene pool and the environment (and remember the environment includes ALL the other organisms gene pools).
If one is to consider local minima (or maxima depending on the measure used) of evolution. To take contrived examples changing bone structure from a human to a bird (a very unlikely path, though if one accepts time reversibility [and given the thermodynamic consideration of evolution this is quite possibly not 100% kosher], evolution, and speciation ther IS a path of living genetic intermediates between ANY two organisms [including you and me]). Obviously this was not a straight geometric morph. At all points it had to be a body structure that could protect itself and function to feed and support itself. The path is hard to imagine and in truth they could be entangled such that it could never be linearized. Especially if you think about them all at once. With spatial separation that allows potentiations and gross overall differences that make adding a new organism to the ecosystem is calamitous.
I have to agree about us not understanding it, but that's not to say we don't understand any of it. Arguing where exactly we are on that continuum is an exercise in futility and much less fun than trying to figure it out. I like it when people toss out ideas rather than facts. Evolution a bunch of "ideas" bounced of the world and eachother for millennia.
Couldn't say it any other way.