In many blind watermarking proposals, the unwatermarked host data is viewed as unavoidable interference. Recently, however, it has been shown that blind watermarking corresponds to communication with side information (i.e., the host data) at the encoder. For a Gaussian host data and Gaussian channel, Costa showed that blind watermarking can theoretically eliminate all interference from the host data. Our previous work presented a practical blind watermarking scheme based on Costa's idea and called 'scalar Costa scheme' (SCS). SCS watermarking was analyzed theoretically and initial experimental results were presented. This paper discusses further practical implications when implementing SCS. We focus on the following three topics: (A) high-rate watermarking, (B) low-rate watermarking, and (C) restrictions due to finite codeword lengths. For (A), coded modulation is applied for a rate of 1 watermark bit per host-data element, which is interesting for information-hiding applications. For (B), low rates can be achieved either by repeating watermark bits or by projecting them in a random direction in signal space spread-transform SCS). We show that spread-transform SCS watermarking performs better than SCS watermarking with repetition coding. For (C), Gallager's random-coding exponent is used to analyze the influence of codeword length on SCS performance.