Intermittently slow CPU performance in AWS Lambda

Consider the following AWS Lambda function:

import time
import pickle
import gc

def ms_now():
    return int(time.time_ns() / 1000000)

class Timer():
    def __init__(self):
        self.start = ms_now()

    def stop(self):
        return ms_now() - self.start

with open('model_embeddings.pkl', 'rb') as file:
    model_embeddings = pickle.load(file)

def get_embeddings(texts):
    timer = Timer()
    embeddings = model_embeddings.encode(texts)   # The line of interest
    print(f"Time: {timer.stop()}ms")
    return embeddings

def lambda_handler(event, _):
    gc.disable()                                  # Disable garbage collection
    result = get_embeddings(event['texts']).tolist()
    return {
        'statusCode': 200,
        'headers': {
            'Content-Type': 'application/json'
        },
        'result': result[0][0], 
    }

The function is deployed as image and is given 10,240 MB of RAM, of which only 800 MB is used. The function outputs the timing of a single line that performs a CPU-intensive task of running input strings through a neural network (HuggingFace sentence transformer) that has been initialized and loaded into memory.

I run this function repeatedly with the same input from AWS Lambda console. The output time is usually around 350ms. However, some runs are as slow as 2100ms. Note that these spikes cannot be explained by either cold starts (since only one line's performance is being measured) or garbage collection kicking in (since garbage collection is disabled).

EDIT: I ran it locally (under WSL under Windows) and never got such a high spike. So it must be something AWS Lambda - specific...

What can cause such spikes?

P.S. The question at SO