Two proxy exits, each with spare bandwidth, look like an easy way to double a download's speed. A normal proxy pool does not work that way.

A client opens a connection through one selected proxy. That proxy opens the onward connection. Adding another address from a proxy list to the pool gives future connections another choice; it does not splice the second proxy into the byte stream already in progress.

Ordinary connection-based multi-WAN balancing has a similar limitation. It can spread different connections across links, improving aggregate capacity or reducing contention. One connection remains one connection unless another mechanism explicitly provides multiple paths.

When parallel downloads are a different proposition

A client may request distinct ranges of a file over separate connections and assemble them locally. That can use more than one route if the client supports the arrangement and the server permits the requests.

It introduces requirements: the ranges must describe the same representation, the server's replies must match the requested offsets, and changes to the object must not produce a mixed file. Authentication, server limits, and per-exit policies can also make parallel range requests unsuitable. Ten complete downloads are not ten pieces of one download.

The final path to the client still matters. Several fast exits cannot send more data through one saturated access link than that link can carry.

Multipath needs actual support

Multipath TCP is an explicit transport mechanism that can carry a connection using multiple subflows when its endpoints and network arrangement support it. It is not enabled merely by listing two unrelated HTTP or SOCKS proxies.

For a slow transfer, first identify whether the ceiling is on the client link, the proxy, the onward route, or the destination. A pool is useful for connection distribution. Treat a claim about doubling one file's speed as a separate feature that needs its own mechanism and evidence.

Mikanda mikatakata