This is the result of somebody not doing his homework properly.

The LFSR outcome is to be used after a full shuffle!!! I.e. if you want 16 bits, you must do 16 "rotations" before you produce an output.
I hope you verified this arrangement produces a full 65536 run length. However, for 16-bit output I'd recommend you at least a 32-bit LFSR. Search the net for recommended configurations, or pester Jez to tell you some :-)

You might want to keep the shift register in the internal memory and copy only the result to the ports.

Other remarks:

• read Jon's "noob" document (btw., english speakers, what the #$%^ is "noob"?), mind the <pre> and </pre> tags and the remark on posting WORKING code (yours now has two errors which prevents assembling).
  
• I don't believe your assembler doesn't support cpl c; but you certainly will find some nicer way for the xor. The presence of bit manipulation instructions does not force you to use them.
  
• calculate the delay - it's not 250ms nor 500ms as your comment says. This is your homework for today :-)
  

Jan Waclawek