Assuming you want the best solution to the problem then the way to parse a data file that contains records separated by some delimiter ie two newlines (aka a blank line) is to set the input record separator to that delimiter and then read the data in one record at a time. For example in your case you could parse the data into a hash like this:

#!/usr/bin/perl
local $/ = "\n\n";
while(<DATA>) {
    next unless m/U\d+/;
    chomp;
    chomp(my $text = <DATA>);
    $h{$_} = $text;
}
use Data::Dumper;
print Dumper \%h;
__DATA__
U448.1

Several walls of the imperial city of Constantinople which had been fr
+eshly
rebuilt with masonry, and fifty-seven towers collapsed as a result of 
+a
violent earthquake which prevailed in various places.

U483.2

Or, this year the battle of Ocha according to others, won by Muirchert
+ach
Mac Erca and by Fergus Cerball son of Conall of Cremthann, and by Fiac
+hra
Lon son of the king of Dal nAraide.
[download]

An inefficient regex solution is:

my $str = join '', <DATA>;
%bits = $str =~ m/^(U[\d\.]+)\n\n(.*?)\n\n/msg;
use Data::Dumper;
print Dumper \%bits;
__DATA__
[snip]
[download]

You need .*? /m /s and /g to make it work and it is not efficient due to backtracking. See perlre