Inconsistent timing results

[tyronerees]

Running the same test, on the same dedicated VM, two days apart gives the following runtime tables:

At the moment, there is too much noise in the times to make any judgement from these tables. In order to improve this, we should:

• run the benchmarks a number of times (five by default, but make this a control parameter), and discard the highest and lowest¹
• consider using the timeit module, instead of time.clock, as it's claimed to be more robust at benchmarking, especially across platforms. This will do the averaging in the previous point naturally see docs.

¹ some minimizers, e.g., the Differential Evolution (de ) minimizer in bumps, will be stochastic, and will naturally give different runtimes. This should be noted somewhere in the table.

[tyronerees]

Needs #263 to be addressed before fixing