Care to show us some sample data and how you'd like the result to look?

Note that you can use (?:...) to group stuff without capturing which can clean things up quite a bit. As a hint, I've added a little white space to your regex to make the groupings more obvious and added a digit at the start of each capture group. Maybe those numbers are not quite what you expect?

s/((\n) ([^0-9])+ (-)* (Aa-Zz)*) | ((\n) (\d{3}) (-)* (Aa-Zz)*)/$2$3/g
+x;
# 12    3         4    5           67    8       9    0
[download]

Update:

Maybe what you want to achieve is something like this:

use strict;
use warnings;

my $wholeBallOfWax = do {local $/; <DATA>};
my @records = split /(?<=\n)(?=\d+-)/, $wholeBallOfWax;

s/\n+$/\n/s for @records;

print join "---\n", @records;

__DATA__
1-12
last non-blank field


2-10
data
more data

3-21
stuff
more stuff
Lots of stuff
so much stuff there is no following empty field
4-73

Sneeky record with a blank field in the middle!

5-00
Last record
[download]

Which prints:

1-12
last non-blank field
---
2-10
data
more data
---
3-21
stuff
more stuff
Lots of stuff
so much stuff there is no following empty field
---
4-73

Sneeky record with a blank field in the middle!
---
5-00
Last record
[download]

In the split regex there is a look behind ((?<=\n)) which matches a new line before the current search point, and a look ahead ((?=\d+-)) which matches one or more digits followed by a hyphen. Neither match "consumes" the string that was matched so the split doesn't drop any characters.

As an aside, the do {local $/; <DATA>} bit suspends end of line detection and reads everything from <DATA> into $wholeBallOfWax (although maybe that was obvious?).