A conventional coil builds its magnetic field at a parabolic rate, but that magnetic field collapses at an inverse parabolic rate (ie: it collapses from 100% faster than it took to build to 100%).

We can take advantage of the disparity between the time it takes to build the magnetic field, and the time it takes that magnetic field to collapse to push the rotor, rather than slow it down.
From the graphic above, we see that the magnetic field of an inductor builds to ~63.2% of maximum in 1 L/R time period, whereas it collapses to only ~36.8% of maximum in that same time period.
Ideally what we want to do is exploit the difference in time and energy between building the coil's magnetic field from ~63.2% to 100% (which would take 4 L/R time periods), and the time and energy necessary to collapse that magnetic field from 100% to ~36.8% (which would take 1 L/R time period). So ideally we'd be putting in 4 L/R time periods to increase the magnetic field by ~36.8% (from ~63.2% to 100%), then utilizing that magnetic field as it collapses in 1 L/R time period by ~63.2% (from 100% to ~36.8%).
Now, obviously that's not going to work for very long... you can't build up by ~36.8% (to 100%) and collapse by ~63.2% for very long before you're hitting 0% during the collapse. The coil on the next few go-rounds would work its way downward in energy until it only builds up from 0% to ~63.2% (because of the rotational speed which gets us that 1 L/R time period). The coil's magnetic field can't collapse below zero, and we can't easily vary the speed of the machine quickly while it's running to take advantage of the 4 L/R time period of building from ~63.2% to 100% (~36.8%), then extracting ~63.2% (100% to ~36.8%) of it in 1 L/R time period, so what we're actually doing when the machine is running is building that magnetic field in the coil to ~63.2% in 1 L/R time period, then collapsing it to 0% in 1 L/R time period before it can act as bEMF. IOW, we've found a way to get around the "hysteresis curve" of the coil.
From the video above, for a normal generator:
L = 106 mH
R
int = 21.1 Ohm
R
load = 200 Ohm
R
total = R
int + R
loadL/R
total = 0.4794 msec
(60000 msec per minute / msec per interaction) / Number Of Poles = RPM
So assuming a 12 pole generator, a conventional generator with the above attributes would have to spin at 10,429 RPM to lessen rotor drag due to bEMF.
From the video above, for a delayed-Lenz generator:
L = 2182 mH
R
int = 384.5 Ohm
R
load = 200 Ohm
R
total = R
int + R
loadL/R
total = 3.733 msec
(60000 msec per minute / msec per interaction) / Number Of Poles = RPM
So assuming a 12 pole generator, a delayed-Lenz generator with the above attributes would have to spin at 1339 RPM to lessen rotor drag due to bEMF.