Just to throw another spanner in the works.
2011/03/14 (00:00:00 server time) to 2011/03/18 (22:59:59 server time)
Is that 5 days or 4.958?
How about
2011/03/18 (00:00:00 server time) to 2011/03/18 (22:59:59 server time)
Or
2011/03/17 (23:00:00 server time) to 2011/03/18 (22:59:59 server time)
On technicality, the first should be 1.00 as it spans the whole of friday, the second should be 1.00 as it covers 24 hours, which is 1 day but should also be 1.0417 which is the whole of friday plus 1 hour.
Now consider
2011/03/17 (12:00:00 server time) to 2011/03/18 (11:59:59 server time)
Again, technically 1 day (24 hours), but technicaly not as you have 0.5 from 03/17 plus 0.5217 from 03/18 which doesn't add up to 1
Back to your original example,
2011/03/11 (11:00:00 server time) to 2011/03/14 (06:00:00 server time)
(12 hours from a 23 hour day = 0.5217) + (6 hours from 24 hour day =0.25) = 0.7717 days or 0.75 days?
Seems to be a very bad idea to work with decimal time when your days are inconsisant in length, whichever way you go with this you will inevitably end up with "incorrect" results somewhere.