This is a complex problem to lay out and depends on multiple factors. Some of these are as follows:
1) Processor: It is important to note that in general two CPUs under comparison are not expected to perform at the same level exactly - for two machines with the same processor there is always a range of acceptable bench results due to process variations and for different processors there is no direct answer for why one might be better than other.
The following website can be used as a reference for a basic comparison between different CPUs on the market:
2) Power mode: In some cases, using high power mode as opposed to balanced or power saver results in an improvement in the results of 'bench' in terms of speed. In particular, this change may lead to improvements in times for the LU, FFT, ODE and Sparse tests.
3) Graphics with Remote access: As far as the 2D and 3D graphics results, it is important to note that using remote access will consume a non-negligible amount of system resources and it is difficult to understand performance impact. Another reason 'bench' results might be slower is because the system can no longer use hardware acceleration to improve graphics performance and is forced to use a software version which is slower than a local implementation.
Running graphics intense programs while using remote access will typically not work as fast as a local implementation of MATLAB. You can further check GPUs specs against the machines used for benchmarking using the following link:
4) Other reasons: Lastly, it is possible that the slowdown is not specific to MATLAB. You can use diagnostic tools like the Intel Processor Diagnostic tool to check the health of the processor and verify if there are other reasons for the system to be slowing down:
