I guess that the problem is with the search and not with the evaluation.
On the other hand, this is a very common problem in chess programs. Almost every strong chess program will announce a mate in N at some point and then switch to mate in N+k, for some k larger than zero from time to time.
That is probably what we should expect from a program that is not a mate solver. All eval and search results are really approximations, and with advanced search techniques available now, giant clumps of the tree are ignored.
I would not single out Junior as buggy (any more than Critter, etc.) and I further suspect that if Junior were allowed to analyze for a full hour it would get the correct depth result. I have some versions of Junior, but not that one, so I can't check it myself.
