I'm not sure users of a no longer maintained module are entirely 'out in the cold'. OK, they're stuck at a certain point, but as long as you don't delete your original file, they face a choice:

1) Update their dependent code to reflect any interface changes with the new merged module

2) Stick with the old module (risking new versions of other modules or perl deprecating some of the code)

In the latter case, we have to assume that an experimental module isn't likely to be a huge complex undertaking like DBI, and a few code fixes to get around deprecated useages are not too much of a problem.

I do think you are right and that users should be clearly warned that the code is experimental though. If it was stated in the POD, readme and filename that would be pretty clear.

I like the underscore idea as well - forcing a user to follow an extra step in installation highlights the experimental nature of the code quite nicely and reduces the likelihood that casual users will inadvertently install it.