In the question "How does the system sustain regular low-effort "splitting attacks"?" a scenario about the ability to disrupt the network by spamming double-spends and having the network work to resolve these incompatibilities is drawn. The question tries to evaluate the possibility to shut the network down by having it resolve so many double-spends that it's basically unable to confirm anything anymore since it has to discard so many half confirmed parts of the tangle that it doesn't get to grow a confirmed part anymore.
That is my question, is there a way to determine how fast a double spend is detected by a IOTA network?
I'm saying a IOTA network because I assume it to be vastly different depending on the amount of nodes, their connection pattern and the load the network has to process. The IOTA whitepaper already makes an important distinction between a high load and a low load regime (v1.3, p.10). I figure it makes quite an important difference how the network topology is made up. Imagine a line of a thousand nodes with only one left and right neighbor each versus a thousand nodes with nine neighbors in a random distribution. Or imagine that distribution being optimized for a low hop count?
How quickly is a double-spend detected and what are the most important parameters to evaluate that speed?