It is impossible to answer this question, because there are different i3 and i5 processors:
Core i5-4302Y, 2 cores, 1.6 GHz
Core i5-2550K, 4 cores, 3.4 GHz
and similar differences for i3's also. You can find an certain i3 which is much faster than a certain i5, and the other way around also.
If one processor has a 5% better performance, this can be measured. But improving the Matlab code for a 10% speedup is possible in any program I've seen yet. The question is only, if the investment for the time to improve the code is efficient: Spending a week to optimize the code is wasted, if the total runtime is improved by several seconds.
Pre-allocation, columnwise processing of arrays, using smaller data types, avoiding the explicit calculation of the matrix inverse, using cells of arrays instead of arrays of cells (or structs), inplace calculations, moving repeated calculations out of the loops, apply maths manually to simplify the problem, ... These techniques can accelerate a program by a factor of 100, while neither a high-power i3, i5, i7 or iXYZ will offer this performance in the next 20 years.
I had an impressive boost in the efficiency of my work, when I upgraded my machine from a Pentium-M laptop to a Core2Duo -- but the main reason was not the computing power, but the fact, that the old Pentium-M laptop had a problem in the fan control and only enabling the sleep mode temporarily decreases the speed of the annoying loud fan after a period of high system load.
The performance can be defined by the time required to solve a problem sufficiently. Then the processor version is only a very small part of the total human-machine system.
