Well, I guess this answers my questions.

This _can_ be done, but...
1) '51 is too "weak" for decent speed. It would need some extra hardware to do it too. If ever, I should think of a stronger processor.
2) Yeah, I've seen Postscript reference book. About 1000 pages. Despite pretty interpreter-friendly syntax, implementing whole Postscript is a lot of work.
3) It won't pay in terms of cost of development. 1 man-year may be not enough. Especially with different printer backends.