When a function that expects to return an IEEE 32ΓÇôbit result finds it out of range, it returns an IEEE 32ΓÇôbit infinity, not an IEEE 80ΓÇôbit number (Exceptions: "gcd" and "lcm".) Thus flonums become 80ΓÇôbit by contagion, but rarely by evolution.
Certain operations can be misleading when different types of numbers are involved: Thus
(= 2e18 #l2e18) => #f
because "#l2e18" is exactly 2000000000000000000, whereas 2e18 is merely the best approximation to it that will fit into an IEEE 32ΓÇôbit flonum.