Drawing from specific needs in physics and in machine learning, we review software engineering systems associated with a selection of Scheme implementations and dialects relevant for scientific computing: Chez Scheme, CHICKEN, Gambit and Racket. We address the needs of an \emph{impatient schemer} who aims at profiting from currently available Scheme systems to solve research and data analysis problems. We examine aspects related to runtime performance, development tools (including availability of external libraries) and parallel computing. Based on two case studies, we first discuss the gap between the few Scheme numerical libraries and the extensive resources available for mainstream languages, which is a serious obstacle for prompt adoption of Scheme systems. Then we suggest that research projects building upon simple components can take advantage from Scheme-based languages to overcome expressiveness and efficiency limitations leading to cumbersome engineering practices in modern scientific computing. Further development of basic Scheme numerical libraries serving as common ground for the advancement of more specialized applications is desirable.