Thx for your extensive response. Yes, now the images do reflect the current design. I found some hours this evening to dive into possible performance improvements. Found something in lit-element about the shouldUpdate() function that could help.
What’s happening? HA gives each custom card a large hass property. Cards check this hass object for changes in entities. By default HA and Lit-element cards are designed to update each visible card on every change of every entity (= automations, sensors, scripts, etc, etc). For large HA setups with many entities this can become a problem if you combine it with one or more of the heavier kind of cards. 
I made a change to the power-wheel-card to update the card only on specific entity changes (e.g. all the input sensors). That should skip a lot of unnecessary updates and should free some browser resources. There’s a version 0.0.9-dev available on the dev branch in the repo if you want to give it a try.
But since HA updates the visible cards only, you could also/first try to temporarily isolate each heavier kind of card on a separate view in Lovelace to test things. This approach could help to pinpoint the performance problems to a specific card.
I hope the dev-release and the test approach can in the end solve some of your performance problems, if not all. 