bunny:/tmp spiros$ perl u.pl 
SV = PV(0x1801460) at 0x180bcf0
  REFCNT = 1
  FLAGS = (PADBUSY,PADMY,POK,pPOK,UTF8)
  PV = 0x30f160 "ba\303\261o"\0 [UTF8 "ba\x{f1}o"]
  CUR = 5
  LEN = 6
match
[download]

utf8::upgrade just forces the string to be utf-8 (otherwise it'll probably latin-1 - i.e. 1-byte encoding) in an attempt to match your problem. Since it doesn't reproduce the warning/error you're reporting I must have missed something.

I note there are 2 variables in your code. Are you SURE the problem arises when both variables are "baņo" AND are both utf-8 encoded? In other words, check the Devel::Peek::Dump output from both variables.

"What should it profit a man, if he should win a flame war, yet lose his cool?"

This is the output for both $term and $token just before the error:

TOKEN:
SV = PVMG(0x1c6cca0) at 0x1ab6a28
  REFCNT = 1
  FLAGS = (PADBUSY,PADMY,POK,pPOK,UTF8)
  IV = 0
  NV = 0
  PV = 0x2dd2f80 "ba\303\261o"\0 [UTF8 "ba\x{f1}o"]
  CUR = 5
  LEN = 15
  MAGIC = 0x2dc95f0
    MG_VIRTUAL = &PL_vtbl_utf8
    MG_TYPE = PERL_MAGIC_utf8(w)
    MG_LEN = 4
--------------------
TERM:
SV = PVIV(0x18b8e20) at 0x1ab69c8
  REFCNT = 1
  FLAGS = (PADBUSY,PADMY,POK,pPOK)
  IV = 0
  PV = 0x2dd3110 "ba\303\261o"\0
  CUR = 5
  LEN = 12
--------------------
[download]

It appears $term is not actually UTF-8 encoded when this occurs. Additionally, is it me or does

[UTF8 "ba\x{f1}o"]
[download]

look wrong ? From what i recall, the UTF8 part of Dump should show the actual word, meaning bano (accented n) and not the encoding.