I wonder if there is any performance/compilation difference between using any of the below.

Interesting question, lets find out. I used Benchmark to figure this out. I ran them through at 50,000 iterations, and I changed all the "print" statements to "warn" statements, so I could pipe STDERR to /dev/null, and still see the results of the benchmark on STDOUT. Here's the code:

#!/usr/bin/perl -w

use strict;
use Benchmark;

use vars qw/ @a @b /; 

@a = ('a'..'z');

timethese(50000, {
   '1' => '@b = map uc,@a; warn @b,"\n"',

   '2' => '@b = map uc $_,@a; warn @b,"\n"',

   '3' => '@b = map uc(),@a; warn @b,"\n"',

   '4' => '@b = map uc($_),@a; warn @b,"\n"',

   '5' => '@b = map {uc} @a; warn @b,"\n"',

   '6' => '@b = map {uc $_} @a; warn @b,"\n"',

   '7' => '@b = map {uc()} @a; warn @b,"\n"',

   '8' => '@b = map {uc($_)} @a; warn @b,"\n"',

});
[download]

And here are the results:

 Benchmark: timing 50000 iterations of 1, 2, 3, 4, 5, 6, 7, 8...
1:  4 wallclock secs ( 4.55 usr +  0.03 sys =  4.58 CPU) @ 10917.03/s 
+(n=50000)
2:  5 wallclock secs ( 4.51 usr +  0.03 sys =  4.54 CPU) @ 11013.22/s 
+(n=50000)
3:  5 wallclock secs ( 4.53 usr +  0.03 sys =  4.56 CPU) @ 10964.91/s 
+(n=50000)
4:  4 wallclock secs ( 4.54 usr +  0.03 sys =  4.57 CPU) @ 10940.92/s 
+(n=50000)
5:  5 wallclock secs ( 4.52 usr +  0.06 sys =  4.58 CPU) @ 10917.03/s 
+(n=50000)
6:  5 wallclock secs ( 4.52 usr +  0.00 sys =  4.52 CPU) @ 11061.95/s 
+(n=50000)
7:  4 wallclock secs ( 4.34 usr +  0.01 sys =  4.35 CPU) @ 11494.25/s 
+(n=50000)
8:  5 wallclock secs ( 4.31 usr +  0.04 sys =  4.35 CPU) @ 11494.25/s 
+(n=50000)
[download]

While these results show #7 and #8 as being the fastest, I would consider this inconclusive. The fact is, it's so close that every time I run this test, I keep getting wide ranging answers, even with 50,000 iterations.

Either way, it was still fun :-)
-Eric

Updated: Fixed a booboo pointed out by chipmunk. The benchmark actually runs correctly now :-)