Hey Christopher, I think i can give my 2 cents on maxwell, as i have been on its beta as well a few years back. This is from what was going on then. I cannot say anything about the current state of the engine as i have not touched it since. Purely from a rendering standpoint, maxwell felt slow, first and foremost because it is an unbiased engine, and it does not cheat its solution. That means in order to get rid of the sampling it needs to do a ton of passes to get an accurate convergence. What that meant for me, as an individual, was that animation was out of the question unless i was willing to work with a grainy image or if i chose to wait a long time for the frames to be rendered. Most people these days rely on farms to render with maxwell in an animation environment (rendernet.se comes to mind). This was the low side of it, and i hear it is quite similar to arnold from this standpoint (good quality takes more samples which in turn takes a longer time to achieve). This is because both engines do not precompute or cache anything. Brute force is the word here, whereas vray, even if it does brute force well, it has a ton of other choices to "cheat" its way through, resulting in a faster rendertime, which in turn, unfortunately, requires greater knowledge from the user.
On the upside, the shading system was nice, had the usual ubershader approach, tons of shaders available in the community. Did not use light sources, but instead turned objects into emitters using a special shader. That meant the shadows and everything else looked very realistic. Its preview system was way ahead of anything at that time in terms of seeing the final look of the image, in the first pass, so you could get a very good idea if you needed to adjust things before waiting for 2 hours. Now this has been updated to the maxwell "fire" engine. But most renderers today give you this (modo's preview or vray's light cache come to mind). By far the most useful feature of the engine for me, was its mxi image format (similar to a raw file), which stored lighting information from all the light sources. That meant if you had screwed up your exposure, lights etc, you could fix everything afterwards, and i don't mean brightness/contrast fix. You could dial the lights in and out, change their intensity, etc, and everything would update realtime in it's "image editor". I hear now they have a nuke plugin for this. Worked for sequences of frames as well, and was a lifesaver. I remember this one time i had an interior to render for a client, and it had around 50 lights total. The guy did a dozen variations, changing colors and turning lights on and off. Had it not been for this feature, i would have been rendering a week on the project. With it, i just waited a couple of hours, and then did a dozen variations in half an hour from the same render. Final thing i'd like to point out, was that its xsi integration was not that good nor stable back then. Maybe now things have changed, but last i looked, it was pretty much the same workflow. Cheers, O