Bitset nature of Erlang guards

What would you say if you'd see following code:

-define(IS_ALPHA(C),
  (C =:= $a) or (C =:= $b) or (C =:= $c) or (C =:= $d) or (C =:= $e) or
  (C =:= $f) or (C =:= $g) or (C =:= $h) or (C =:= $i) or (C =:= $j) or
  (C =:= $k) or (C =:= $l) or (C =:= $m) or (C =:= $n) or (C =:= $o) or
  (C =:= $p) or (C =:= $q) or (C =:= $r) or (C =:= $s) or (C =:= $t) or
  (C =:= $u) or (C =:= $v) or (C =:= $w) or (C =:= $x) or (C =:= $y) or
  (C =:= $z) or
  (C =:= $A) or (C =:= $B) or (C =:= $C) or (C =:= $D) or (C =:= $E) or
  (C =:= $F) or (C =:= $G) or (C =:= $H) or (C =:= $I) or (C =:= $J) or
  (C =:= $K) or (C =:= $L) or (C =:= $M) or (C =:= $N) or (C =:= $O) or
  (C =:= $P) or (C =:= $Q) or (C =:= $R) or (C =:= $S) or (C =:= $T) or
  (C =:= $U) or (C =:= $V) or (C =:= $W) or (C =:= $X) or (C =:= $Y) or
  (C =:= $Z)
).

The first thought comes in the mind is to replace this macro with one-liner:

(X > $a andalso X < $z) orelse (X > $A andalso X < $Z)

And really when you put this macro in the body

is_alpha_cow(X) -> ?IS_ALPHA(X).
is_alpha_bull(X) -> (X >= $a andalso X =< $z) orelse (X >= $A andalso X =< $Z).

you receive predicted numbers:

1> sample:mbull().
{119,<<>>}
2> sample::mcow().
{1598,<<>>}

Having measuring functions as follows:

mcow() -> Bin = crypto:rand_bytes(1000), timer:tc(sample,measure_cow,[Bin]).
mbull() -> Bin = crypto:rand_bytes(1000), timer:tc(sample,measure_bull,[Bin]).
measure_cow(<<>>) -> <<>>;
measure_cow(<>) -> is_alpha_cow(B), measure_cow(Bin).
measure_bull(<<>>) -> <<>>;
measure_bull(<>) -> is_alpha_bull(B), measure_bull(Bin).

However when you put macro inside the guards:

is_alpha_cow(X) when ?IS_ALPHA(X) -> true;
is_alpha_cow(X) -> false.
is_alpha_bull(X) when (X >= $a andalso X =< $z) orelse (X >= $A andalso X =<
$Z) -> true;
is_alpha_bull(X) -> false.

you receive not only better result than in body, but also the best result of
[{guard,body}x{cow,bull}] measurement:

3> sample:mbull().
{104,<<>>}
4> sample:mcow().
{64,<<>>}