Will switching to server side experiments improve performance and eliminate the bias introduced by client side experiments?

Question

We have a server site rendering React application and currently are using client site A/B testing tool. The issue here is following:

1. User comes to the site
2. Server loads default version
3. Client detects if user should be in varation 4.If Variation: Start from 2 again
4. Output

So as you can see, the variation is always slower.

I would like to know a good an industry standard way, how to solve this problem? Should we switch to server side experiments?