It's equivalence: returning boolean makes sense. And it mirrors the VarEqv documentation.
https://msdn.microsoft.com/en-us/library/windows/desktop/ms221077(v=vs.85).aspx

What's the problem?

$a = new VARIANT(255, VT_UI1); // 11111111
$b = new VARIANT(2, VT_UI1);    // 00000010
$c = variant_eqv($a, $b);

echo "result is: ".$c;

result is: 2     <--- Does not match description   "If each bit in left is equal to the corresponding bit in right then TRUE is returned, otherwise FALSE is returned."

It is bitwise, not logical, operation.

> It is bitwise, not logical, operation.

Indeed, so apparently MSDN is wrong, also wrt. VarXor which also
works bitwise.

After some testing, VarEqv appears to return the same/similar type of value as it was passed; give it VT_BOOLs and it returns a VT_BOOL, or give it VT_UI8s and it returns a VT_UI8. I've also found some anecdotal evidence that VARIANTs have always blurred the line between integer and boolean.
Presumably this behavior extends to any situation where integers and booleans can be treated the same way, though with a bit more testing the return type is the same boolean/integer nature but not necessarily the same VT_* type - for example, VT_UI8(255) + VT_UI8(2) = VT_I4(2).

I'm thinking a warning in the variant_* functions that work with booleans (probably and,eqv,imp,not,or,xor), and/or a note in the VARIANT class docs that this type looseness happens.

> VT_UI8(255) + VT_UI8(2) = VT_I4(2).
Er, that's an AND, not addition.

It seems to me the COM documentation needs a general overhaul.
There's still a lot of talk about the differences between PHP 4
and 5, which is water under the bridge now.  Also, the return type
of variant_eqv() and lots of other variant_*() functions is given
as `mixed`, but is most certainly `VARIANT`.