MATLAB/Python Engine Integration Optimizing for Execution Speed

I have a function in MATLAB that I've timed to run at about 0.25 seconds per iteration using tic/toc. However, because of integration with other systems, I'm trying to call that function from Python using the MATLAB Engine by passing in a few input parameters: namely three vectors around ~5000, ~5000, and ~60000 elements (the return is a single double value):

eng.runFxn(obj, matlab.double(v1), matlab.double(v2), matlab.double(v3), float(i + 1))

However, this has been slower by an order of magnitude, running at around 3-4 seconds per iteration (using time.time()). I'm just wondering if this is expected behavior/if there's any way to speed up the inter-language communication speed, or if this approach may be a lost cause and I need to find a different method to execute the MATLAB script.

[Steven Lord]

I have a function in MATLAB that I've timed to run at about 0.25 seconds per iteration using tic/toc.

With the same input arguments that you're passing into the function when you call it from Python?

However, this has been slower by an order of magnitude, running at around 3-4 seconds per iteration (using time.time()).

Where is the time being taken? In the function call itself, in the conversion of the data to double with matlab.double, or somewhere else? If it is the conversion that's the bottleneck, can you perform that conversion prior to the start of the loop where you're iteratively calling the function to limit the number of times you need to convert?

[vyw2]

I wasn't 100% sure myself--the MATLAB function takes around 0.2-0.4 seconds to run:

But the Python one is much slower (with the same arguments):

I did think it was the conversion, but it doesn't seem like the conversion takes up that much time(?):

Which is why I was slightly unsure if I was messing up the implementation, or if the overhead really just does take that long. I glanced at this answer as well:

Why is MATLAB engine for Python so much slower in R2023b vs. R2021b? - MATLAB Answers - MATLAB Central

which implied that casting to a matlab.double may help, but that did not seem to speed up the function execution appreciatively.