What's actually happening
Telegram uses its own protocol, MTProto, and operates datacenters in multiple regions (primarily Frankfurt, Amsterdam, Miami, Singapore, and others). When a signal provider sends a message, it arrives at a Telegram DC almost instantly. From there, your client device has to pull it down.
The speed of that pull depends entirely on the route your device has to the relevant Telegram DC. If your ISP takes a congested or indirect path to that DC, the delivery window grows. Three seconds of additional latency is enough to miss a scalp entry entirely.
This is compounded during peak market hours. At London open and during news events, general internet traffic spikes - which adds congestion to the same backbone links your Telegram traffic uses. The delay gets worse at exactly the moment you need it to be shortest.