Floating point addition, without floats!

Python, 224 chars

This code converts a floating point input f to the integer f*2^150, does the addition using python native big integers, then converts back.

V=lambda x:((-1)**int(x[0])<<int(x[1:9],2))*int('1'+x[9:],2)
B=lambda x,n:B(x/2,n-1)+'01'[x&1]if n else''
def A(x,y):
 v=V(x)+V(y)
 s=e=0
 if v<0:s=1;v=-v
 while v>=1<<24:v/=2;e+=1
 if e>254:v=0
 return'%d'%s+B(e,8)+B(v,23)

J (172 characters)

Since IEEE 754 allows five rounding rules, I chose the "toward 0" rule. Here is my code:

b=:(_1&^)@".@{.*[:#.1x,("."0)@(9&}.),#.@:("."0)@}.@(9&{.)$0:
a=:b@[([:,(<&0)":"0@:,,@((8$2)&#:)@(-&24)@$@#:,}.@(24&{.)@#:@|)@(]`(**2x^278"_)@.((>&((2x^278)-2x^254))@|))@+b@]

The very same examples you give (but not exactly the same results because of the different rounding rule):

   '01000100011110100000000000000000' a '00111111000000000000000000000000'
01000100011110100010000000000000
   '01111111011111111111111111111111' a '01111111011111111111111111111111'
01111111100000000000000000000000
   '01000011101000001000111110111110' a '11000010111101101010010001011010'
01000011010001011100110101001111