Slow performance with plotly chart builder

Hello all,
I need to use plotly as the backend of a microservice who generates charts dynamically.
Unfortunately, after a little benchmarking, I found that the plotly.express framework is very slow (around 5 secs to generate a chart from 500 lines dataset).

Here is the script I use to generate a scatter matrix:

import sys
import os
import traceback
import json
import time
sys.path.append('c:\\statwolf\\python\packages\Lib\site-packages')
input = json.loads('{\"file\":\"/tmp/data.tsv",\"color\":\"club_country\",\"dimensions\":[\"nolo\",\"tolo\",\"yolo\"]}')
def action():
    def run():
        import plotly.express as px
        from pandas import read_csv
    
        color = None if input['color'] == "" else input['color']
    
        first = time.time()
        d = read_csv(input['file'], sep='\t')
        second = time.time()
        fig = px.scatter_matrix(d, dimensions=input['dimensions'], color=color)
        third = time.time()
        j = fig.to_json()
        fourth = time.time()
    
        print('read: ' + str(second - first))
        print('plot: ' + str(third - second))
        print('json: ' + str(fourth - third))
        return j
    
    import time
    
    for i in range(0, 3):
        start = time.time()
        result = run()
        end = time.time()
    
        print('iteration: ' + str(i) + '\ntime: ' + str(end - start))
    
    return result
    
result = None
try:
    result = { 'outcome': action() }
except Exception as e:
    traceback.print_exc()
    result = { 'error': str(e) }
resultDir = os.path.dirname(os.path.realpath(__file__))
resultFile = open(resultDir + '/result.json', 'w')
json.dump(result, resultFile)
resultFile.close()

from the dataset:
https://www.dropbox.com/s/cm9i3pfv10exbba/data.tsv?dl=1

and this is the report with timing:
https://www.dropbox.com/s/l2x3jqzea4i4xqw/report.txt?dl=1

Now:

1. Is there any tweak I can implement to improve performances?
2. Do you plan to focus on speed for the following releases?

Start by rerunning the supplied benchmark script and reviewing the linked timing report, separating read_csv, plotly.express.scatter_matrix, and fig.to_json timings. Trace the scatter-matrix entry point and existing performance tests or benchmarks; done would require identifying a reproducible bottleneck and agreeing on a measurable improvement, but the issue does not define an acceptance target.