For 27/101, and if I set the value of the MSbit to 1, I get:
0010 0010 0011 0111 1100 0011 0010 1011 0001 0110 1100 1111 11011
which are then followed by binary zeroes. Which implies to me that I've reached the limit of the accuracy of _Float.
This is where I compute the binary numbers, bit by bit. Also, my super duper program also gives the hex equivalent.
hex 2237C32B16CFD800