The modulus operator is strictly defined for integer numbers, because it "... basically just returns the remainder of an integer division operation ..." (paraphrasing from the K&R book)

In this case the floats are truncated to integers and then the modulus operator is applied, i.e.

203.5 turns into 203
17.9 turns into 17

then 203 % 17 = 16 (as expected)

Other languages follow this approach, while other round up the float previous to operating on them, in those languages the result would be: "6"

And other languages, like Java, have a third behavior in which when using floats, a number of integer substractions is made and a floating point reminder is calculated, see for example: http://softwaredev.earthweb.com/multi/article/0,,12079_630791,00.html
(explanation to question 3)

BTW, gawk also performs an integer number of substractions, if you try:

gawk '{ print (203.5 % 17.9) }'

you'll get "6.6"

Bottomline, as this is an operator meant to work with integers, make the appropriate conversion for the appropriate results, or implement a modulus function like:

function modulus_of ($q, $d) {
  $rem = $q;
  while ($rem > $d )
     $rem -= $d;
  return $rem;
}

which *will* return "6.6" for floats, and should also work on integers.