Thoughts...
Some of the things holding this problem in place seem to center around:
1) "Firefighting" mentality
An arbcom (or indeed any body) that is perpetually rushed from case to case will burn out, and will not have time to decide whether it's tackling things as best they could be. There needs to be slack within the arb' system, a 110% load is going to render any long term thought impractical. This is a major cause of problems in business management, and "management" of matters in general, not just Arbcom.
This can in part be addressed by streamlining the arb process, so that cases take less time (by delegation, by having evidence pre-digested, checked and summarized by clerks and trusted admins). It can also be addressed in part by changing the structure of the processes involved (pool of spare arbiters, multiple circuits, looking at where slowdown happens in a case that's not productive). It can also be addressed in part by recognizing time needs to be made, when Arbcom is fresh, to consider measures to minimize the predictable summer slowdown, rather than walking into it, and tracking the situation and considering remedies (stepping outside the problem to view it from outside). These and other ideas will all have strengths and weaknesses. But each has the potential to help, and allocating time to their consideration will allow a good balance to be chosen.
2) High expectations by the community
As referenced elsewhere, there is a high level of expectation. The cases are more complex, often highly multiparty; the evidence is complex; arbitrators have outside lives and pressures, and yet are expected to rule on these complex cases rapidly and in a way that overcomes arbcom's own differing views and satisfies everybody elses varying feelings. There is also considerable work "behind the scenes" (emails and the like, reviews, requests that nobody will take rulings on except from Arbcom, appeals, etc) that most people don't see. The level of expectation translates into a heavy duty job, that can take Wikipedians and burn them out, or expose them to a never ending treadmill. It's not easy, and those willing to try deserve respect.
3) Lack of communal coherence
The community size is such that decisions on process change are not easy to gain consensus. RFA is one perennial example. The situation that everyone can see problems but no agreement can be reached on change (until it is so dire that everyone will agree on * something *) is not implausible.
Against this, arbcom has an advantage that it essentially can set its own processes, answerable to the judgement of its own members, within a fairly wide remit. The problems that have dogged consensus reaching on other major processes may be less an issue in this arena. It is probably necessary for Arbcom to do this, since it's unlikely the community as a whole will reach broad agreement in any reasonable time frame.
4) Mixed sense amongst administrators what exactly the community has delegated them to do.
I think this is more an issue than is commonly recognized. Wikipedia has a flat structure for "unilateral rulings " - basically administrators (either individually, in discussion, or via many processes), then arbcom. The flip side is that the majority of cases need to be handled (* effectively *) by administrators, because Arbcom is neither designed nor has the capacity or role, to handle more than the very small number of exceptional cases.
In some cases where administrators have not felt "safe" to take firm action, or are worried whether they will have communal support or be criticized and "thrown to the lions", mean that a large number of disputes where clear problematic conduct occurs, are not dealt with at this early stage, but grow and fester. Inevitably more of these end up being more disruptive, drawing more people in, and clogging Arbcom later on. Administrators may need reassurance that they * do * have communal sanction to deal with problem editors and problem disputes at an early stage, or when borderline conduct only, is already visible. Often a small use of tools early on when there is misconduct happening, can keep a dispute to a smaller scale before it gets to be a major mess (eg, the effect of 3RR on revert warring). If admins do not feel they will be supported to fairly but firmly address situations when early stage misconduct is clearly occurring (low level stonewalling, incivility, OR, pov pushing, etc... the kinds of things some admins are reluctant to take action on), then this interferes with the communal delegation of conduct handling to them, and it's likely a proportion of these cases will escalate and inevitably clog up Arbcom a few months down the line.
These four are obviously not all the issues, but taken together, they are clearly matters that help to continue the problems described.
Movement on, or reviews of, these will probably be helpful.
</thoughts>
FT2.