Unfortunately, Perl doesn't optimize [x]:

[~] $ perl -Mre=debug -wle'"x" =~ /x/'
Freeing REx: `,'
Compiling REx `x'
size 3 first at 1
rarest char x at 0
   1: EXACT <x>(3)
   3: END(0)
anchored `x' at 0 (checking anchored isall) minlen 1 
Omitting $` $& $' support.

EXECUTING...

Guessing start of match, REx `x' against `x'...
Found anchored substr `x' at offset 0...
Guessed: match at offset 0
Freeing REx: `x'
[~] $ perl -Mre=debug -wle'"x" =~ /[x]/'
Freeing REx: `,'
Compiling REx `[x]'
size 10 first at 1
   1: ANYOF[x](10)
  10: END(0)
stclass `ANYOF[x]' minlen 1 
Omitting $` $& $' support.

EXECUTING...

Matching REx `[x]' against `x'
  Setting an EVAL scope, savestack=3
   0 <> <x>               |  1:  ANYOF[x]
   1 <x> <>               | 10:  END
Match successful!
Freeing REx: `[x]'
[download]

And in case that doesn't convince you:

[~] $ perl -MBenchmark -we'timethese(-5,{x=>q{"x"=~/x/},"[x]"=>q{"x"=~
+/[x]/}})'
Benchmark: running [x], x, each for at least 5 CPU seconds...
       [x]:  4 wallclock secs ( 5.23 usr +  0.00 sys =  5.23 CPU) @ 85
+2787.57/s (n=4460079)
         x:  6 wallclock secs ( 5.25 usr +  0.00 sys =  5.25 CPU) @ 14
+99188.19/s (n=7870738)
[download]

P.S. I think you missed a backwhack in your RExen above; Perl is going to try to interpolate $flat.