None of the transformers show both second order intercepts at the instrument limit, although the 14 turn transformer wound on a Steward 35T0501 is instrument limited at 1004 KHz.

In general, the transformers wound with 7 turns on a BN73-202 core are excellent performers. However, the best performing transformer is 14 turns bifilar wound on a 35T0501-10H core, with three output intercepts instrument limited and the fourth intercept within 4.5 dB of the instrument floor.

The spectrum analyzer image below shows the 2003 KHz product level for four transformers. As  the plot shows, there's a striking difference between the best and worst transformers.

There is a significant physical difference between the 75 and 35 cores, however. The Fair-Rite 75 core is "unfinished" in the sense that it is uncoated and has only minimal edge smoothing. In contrast, the Steward 35 material core has been tumbled and coated. It has smooth edges and the ferrite body is coated with an insulating material. This physical difference is important because ferrites are hard and abrasive. It's unfortunately all too easy to abrade the enamel insulation on the magnet wire when winding the transformer on an unfinished core.

As I said on this page originally "It's possible that one of the wires has damaged insulation and is coupling electric field energy to the core ... I'll wind another transformer with Kynar insulated wire and see if the same intermodulation performance is observed."

I've done that and the intermodulation performance of the Kynar insulated transformer, ID FT75-02, is excellent. Close observation of FT75-01 revealed several places where bare copper could be seen. At least one of these bare patches must have contacted the core, placing a combined electric and magnetic field in the core.

Interestingly, there is little difference in frequency response between the transformer wound with damaged insulation and the Kynar insulated transformer. The damaged insulation transformer has a bit greater loss at mid-band, but less loss at higher frequencies. The high frequency difference would well be due to insulation dielectric loss.

In any event, it pays to be careful when winding a transformer or inductor on an unfinished ferrite core. Either insulated wire should be used, or the core should be coated or wrapped with thin Teflon tape.

Coilcraft and Mini-Circuits are two manufacturers of wide band RF transformers and I measured the frequency response and intermodulation performance of a small sample from each manufacturer. A few of the measured transformers are shown below, along with one of the home made transformers wound on a BN-73-202 core. In addition, a home made transformer wound on a small core is shown second from the left. The transformers are equipped with 0.1 inch space pin headers for quick connection to test equipment.
 

The 3.007 MHz and 4.011 MHz test signals are adjusted to a level of 0 dBm for each test tone at the transformer's output. For loss loss transformers, the input and output test signals are essentially identical, within a couple tenths of a dB or so. The series connected 1:2 transformers, however, have an appreciable loss, around 1.5 dB. Hence, the transformer on the input side is driven with a signal at +1.5 dBm. In theory, therefore, intermodulation products generated in this transformer of the pair will be greater than those generated in the transformer nearest the output.

The table below presents the output intercepts, 2nd and 3rd order, for the seven Coilcraft and Mini-Circuits transformers.