How about a two-stage approach? First, peel off your units:

   s/(\D+)$//;
   my $units = $1;
[download]

Then proceed with the more standard problem of recognizing a number. See:

• How do I determine whether a scalar is a number/whole/integer/float?
• Checking whether a $var is a number

Normalizing your input in this way can be a powerful trick that allows you to more clearly identify and deal with your edge cases. Unless performance is absolutely stupendously crucial there is no reason to implement a monolithic regexp that handles every possibility.

Matt