The issues surrounding color accuracy can be quite complex. In the case of the Foveon sensors used in the Sigma models mentioned, each pixel location is sensitive to red, green and blue, via layers designed to be sensitive to each of those colors.
The primary benefit of that approach is that you eliminate the color interpolation process you need to perform for Bayer type sensors (where each pixel is only sensitive to one color and you need to look at the values of adjacent pixels in order to "guess" the correct RGB Values to associate with each pixel location during the demosaic process).
So, in theory, that should make the Foveon approach much better. In practice, they've always had some issues with color discrimination between the 3 layers, requiring very powerful software with complex algorithms to convert from raw to a usable image. Now, Sigma's software has gotten *much* better over the years, and you don't see as many complaints in some of the popular forums as you used to about color problems. For example, hue shifts in different lighting (or even in the same image if you had a wider range of bright to dark in it), color shifts as ISO speeds change and more. But, I still see some threads complaining about color problems (for example, red channel clipping, hue issues with both red and blue depending on lighting conditions and more). In perfect lighting, they can have very high color accuracy at lower ISO speeds. In more complex conditions, it can become increasingly difficult to achieve good color accuracy (although newer models have continued to improve, and Sigma's software has continued to improve).
Even the latest offering (DP2) appears to have a lot of complaints about color casts; sometimes only in image corners (my guess is that it might be due to a bit of vignetting from the lens, where hue is shifting because the corners are slighter darker). But, newer software from Sigma will probably get a handle on it, as they continue to refine the latest Sigma Photo Pro offering for it. I can imagine the algorithms for determining what values to assign to red, green and blue are very complex (despite the lack of need for color interpolation from multiple pixel locations like you need to do with Bayer pattern sensors). Dave Coffin (the author of dcraw.c
) even excluded the foveon sensor processing from being licensed as open source (even though all other portions of the code are open source), due to the complexity of the algorithms to try and get accurate colors at each pixel location in different lighting conditions, mostly because of poor color separation between sensor layers from comments I've seen from Dave (in other words, it's so hard to do, he withheld that portion of code in order to make a profit on it for commercial products wanting to use it).
Sigma owners tend to be quite passionate about the color from their cameras. You'll have to make up your own mind, as they tend to be hotly debated cameras.
Unfortunately, given your budget, the Sigma cameras mentioned by TCav are going to be out of your price range, even if you decide they're a good match for what you want to shoot, in the lighting you want to shoot in.