Answering my own question.
I've discovered why the "continuous VX futures" series above has vastly less decay than the nominally equivalent VIX ETF(s).
The standard way in which continuous futures series are generated is plain wrong! The chart above is from TradingView. It shows (the lower line) a "continuity adjusted" price series generated from the individual front-month contracts, but the way that the discontinuity at rollover is removed is by using an additive constant applied to all prior data. The constant is the difference between close of the expiring contract and open of the following month's contract; at each rollover such a constant is applied to all points preceding the rollover.
This is clearly a wrong methodology. But I've discovered that the continuity-adjusted futures data I have purchased from a vendor (apparently unrelated to TradingView's) uses the same methodology - indicating it appears to be a standard method.
The correct way to back-adjust prior data points to remove the discontinuity is with a multiplicative constant (with each rollover, multiply all prior data points by the ratio of prices at the instant of rollover).
(The "standard" methodology is in fact doubly wrong: it generates the additive constant by differencing close of the expired contract and following day's open price in the next month's contract, thus erroneously including any gapping from close to next day's open. Of course, one will ordinarily roll over from one contract to another in very quick succession.)
The multiplicative method is correct because to maintain continuous exposure (/leverage) to the new contract, one needs to buy n1 contracts where n1 = n0 * p1/p0 (where n0 the number of contracts in the expired future, and p0 and p1 are the respective prices of old/new contracts). In fact I cannot see any reasonable way one would replicate the exposure represented in the additively adjusted series, as to do so (for a decaying future like VX) would require that one continually increases one's exposure(/leverage) over time.
The incorrect methodology also explains why the so-generated continuous series for VX appear to be decaying exponentially faster and faster and getting exponentially more volatile.
I can only assume that this standard (and wrong) methodology has stood the test of time because most futures do not continually, reliably, and significantly decay in value from contract to contract. The additive and multiplicative methods will not produce massively different results for most commodities or underlying instruments, but for VX they produce utterly different results (over the long term).
The lesson: for volatility futures, do not use the continuity-adjusted historical series you are likely to get from most sources: they give a thoroughly misleading impression of the real decay characteristics of the futures, and of their internal volatility over time.
I am now generating my own multiplicatively-adjusted series both for front month rollover and further-away rollover. The generated front-month series looks much more like the ETF data that it does the misleadingly additively-adjusted data.