I think it's impossible to get time less than 0,015 sec at new testing system (since approx. end of 2008). Look at problem 1000 - simple operation, but you can't now get time less than 0.015 neither on C++ nor on Pascal.
Furthermore, step between time measurment is about 0,016 (so you can get 0,015 or 0,031 but nothing between)
And furthermore, when you run you some hard solution sequental some times, it will get significant different time result. For example, my Pascal AC solution for 1001 sended a few of times takes 0.078, 0.093 or 0.109 sec
I think it because mistake in time-measurment method. I think they using a system-clock (RTC) that renew after significant time. So, if testing system is hard loaded (for example testing many authors in parallel) the testing will take more time relatively ro Real-Time Clock, than when system is idle (at early morning, for example).
So, testing system is incorrect and you can't reach time less than 0.015 until realisation of system will be changed