The design of the Nx586 had begun in the days of the 386. The separate coprocessor was the most normal thing in the CPU business back then. But things changed and by the time Intel released the Pentium, everybody had forgotten about those days.
While NexGen's CPU performed very well against the Pentium, its roots in the old were very visible : the coprocessor solution was a separate chip!
The Nx587 was designed as a separate chip because of design complexities and limitations of manufacturing processes. Its development was very difficult and problems in the design phase resulted in many delays.
NexGen wanted to focus on the development of the CPU and outsourced the development of the FPU to Chips & Technologies. Unfortunately C&T could not deliver a FPU in the agreed timeframe and NexGen took the development back in-house. Only with the greatest effort NexGen completed the design, but the production of the Nx586 had already started. With no time to spare, NexGen launched its Nx586 without the coprocessor option.
NexGen put the Nx587 in production but by the time the first FPU's were ready, the situation had changed drastically. Many Nx586 chips and motherboards had already been shipped. NexGen's resources were stretched to the maximum and it could not afford to diversify its productline with new motherboards with coprocessor sockets. NexGen decided that the separate FPU would not be released. Instead it continued to sell the coprocessor-less Nx586 and launched the Nx586FP. This chip was packaged in a Multi Chip Module which housed a separate CPU and separate FPU.
When AMD bought NexGen it used the Nx587 in its next generation CPU, the K6. While integer performance of the K6 was more than a math for Intel's Pentium, the floating point performance was just not good enough to beat it. The reason was, as we know now, that the K6 had a coprocessor designed for a CPU generation back !