Still remember when working on an ICL 1901T, the EDS8's (8kb disc storage) were replaced by EDS100's (100kb). Quote "We'll never fill one of those." 
My first actual job involved a 1901T, with 3 x EDS60 drives and 24 kilowords of RAM. There was physically 32Kwords, but they wouldn't pay for the extra.
We used to run 3 batch streams most of the time, though EXEC could keep track of 64 tasks if there was enough memory to hold them in.
The real techies in 1900 kit knew the significance of a magic number - 7036875.
The "official" method of converting binary to decimal consisted of dividing the integer by a power of 10 - if you wanted a 4-digit number, divide by 10000 - then using the CBD instruction to give each digit in turn.
The largest positive integer in a 24-bit word is 8388607. So, to print that, you need to divide by 10000000. Problem! you need double length arithmetic, and that's HARD.
Some clever soul worked out that if you multiply an integer by 7036875 (usually defined as a constant called AMAGIC) you end up with the same bit pattern as if you had divided by 10000000. No double-length sums, and as a bonus, multiplications are faster than divisions. So if you want a 4-digit number, it is quicker to multiply by AMAGIC and throw away the first 3 digits!
