As far as filters for your hyperspectral DLP projector, yes they are readily available. Image Engineering uses filters like this in their camSPECS camera calibrators.
256 bands might be overkill, but I would still like to view an implementation of this concept. If you're going this far, rather than have a bunch of filters, I'd just build a super-bright super-fast sweeping monochrometer (uses a diffraction grating and white light to make any wavelength you want). As far as human perception, there will be diminishing returns as one goes from 3 primaries to 4 primaries (like Sharp's Quattron display with RGBY) and beyond. If the goal is a tool for research into human vision and rendering, then the more the better. I would think you'd be doing pretty good with 30 bands (10-12nm spacing), but you'd need more if you wanted to truly simulate fluorescent lights. It would be nice to have pixels in the near IR and near UV as well. Young children can see much further into the UV than adults. People who've had cataract surgery and opted for non-UV blocking lenses (a friend of mine recently did this) can see to the mid-300nm's.
Here's my gut feel for the number of bands that would be be good for different applications (I'm interested in other folks opinions on this) There is diminishing returns in all of these, and keep in mind that for most applications, existing wide gamut 3-primary displays work really well, if calibrated correctly and used correctly in a color-managed workflow.
More accuracy in soft-proofing prints: 4-6 primaries
Covering PhotoRGB: 5 primaries
Reproducing highly-vivid objects, like neon lights (Neon Tunnel at O'Hare Airport), bird feathers (Peacock, Scarlet Tanager, Scarlet Macaw, Kingfisher, Painted Bunting), butterflies (Emerald Swallowtail, Blue Morpho), certain specialty paints (including fluorescent paints): 5-7 primaries
Color Vision Research: 50-200 primaries (or 'bands') that extend from 350nm to 750nm
If you divvied the primaries up across multiple DLPs and had them all point at the same screen, you could build a hyperspectral display without the temporal artifacts in time-multiplexed color displays. The latter happens as your eye moves across the screen as it refreshes each color. If you have a white line on a black background, you'll see a nice rainbow. It is fatiquing, even in 180Hz displays (60Hz/color in a 3-color system) showing a friendly (non-pathological case) picture.