Blender python iteration of importing model gets slower

Question

I'm doing a project with python in blender and let me summarize its process.

1. Import a .pmx file(kind of 3d model)

       start = time.time()
       bpy.ops.mmd_tools.import_model(
           filepath=str(model_path), 
           files=[{
               "name":os.path.basename(model_path), 
               "name":os.path.basename(model_path)
           }], 
           directory=os.path.dirname(model_path), 
           types={'MESH', 'ARMATURE', 'MORPHS'}, 
           scale=0.08, 
           log_level='WARNING')
       #This import_model function is from addon.
       print("import time :", time.time() - start)
   

2. Analyze and append some data to a global variable 'list[]'

3. Delete every data(Object, Meshes, Materials, Textures, images, texts and collections), and even Purge_all.

   def clean_blender():
   
   for dataset in [bpy.data.objects, bpy.data.meshes,
                   bpy.data.materials, bpy.data.textures, bpy.data.images, bpy.data.texts,
                   bpy.data.collections]:
       for key, value in dataset.items():
           if key != 'camera' and key != 'light' and value.name != 'Text':
               dataset.remove(value, do_unlink = True)
   

   This clean_blender() deletes every data from my scene, and purge_all() delete all orphan datas.

   def purge_all():
       area = bpy.context.area
       old_type = area.type
       area.type = 'OUTLINER'
       bpy.context.space_data.display_mode = 'ORPHAN_DATA'
       bpy.ops.outliner.orphans_purge()
       area.type = old_type
   

   4. Iterate 1 ~ 3 for every 3d models.

But I figured out that each time for "importing" model gets longer, and I checked it via time.time() by iterating just only one file in While 1:. And the time spent on importing(1 Loop) got longer overtime. I've tried many other options

1. No 'For loop', iterate manually like import(file_list[0]) import(file_list[0]) import(file_list[0])...
2. import gc and gc.collect()
3. modify api and make dummy_view_layer_update() to run view_layer_update() only at the end of my script

However It was all same. But the weird thing is that after I exit my script and get back to blender, everything is deleted successfully. And when re - run script, then the speed gets back as fast as the first time I ran my program and gets slower again.

So my conclusion is that Process 3 - 'Delete every data' works fine after the script exits, but not properly during running(Maybe because blender is blocked while running script?) . It is just a guess, and I can't even figure out what should I do.

Or is it just an innate problem of blender python?

Thanks.