Recently, I've been noticing that on-chain data always seems to "pause" for a moment. At first, I thought my internet connection was bad, and anyway, the page kept spinning, almost driving people crazy. Later, I realized that many times it's not the blockchain slow, but the data pathway you're using is queuing: The Subgraph indexer needs to process the new blocks first before sending the data to you, and if there's reorganization or the index is behind, it will pause updates briefly; plus, with RPC rate limiting, during peak times, when requests flood in, you get a 429 error, resulting in your balance/transactions appearing to be half a beat slow.

Now, my own approach is pretty simple: I keep two RPC providers for key pages, separating read and write, lowering concurrency in parameters, increasing cache size—better slow and steady than jittery.
On the L2 side, they compare TPS, fees, and subsidies every day. To put it plainly, front-end experience still depends on how stable these "middle layers" are; otherwise, even if it's fast, it’s like driving fast while stepping on the brakes.